[clang] [dataflow] CXXForRangeStmt should extend flow condition (PR #80989)

[via cfe-commits]

================
@@ -1710,4 +1710,25 @@ TEST_F(TopTest, ForRangeStmtConverges) {
                 // analysis converged.
               });
 }
+
+TEST_F(TopTest, ForRangeStmtHasFlowCondition) {
+  std::string Code = R"(
+    #include <array>
+    void target(bool Foo) {
+      std::array<int, 5> t;
+      for (auto& i : t) {
+        (void)0;
+        /*[[p1]]*/
+      }
+    }
+  )";
+  runDataflow(Code,
+              [](const llvm::StringMap<DataflowAnalysisState<NoopLattice>> &Results,
+                 const AnalysisOutputs &AO) {
+                  ASSERT_THAT(Results.keys(), UnorderedElementsAre("p1"));
+                  const Environment &Env1 = getEnvironmentAtAnnotation(Results, "p1");
+                  ASSERT_TRUE(Env1.proves(Env1.arena().makeAtomRef(Env1.getFlowConditionToken())));
+              });
+}
----------------
martinboehme wrote:

See comments above -- I'm not sure that the test, if it fails, means that we need a flow condition for the range-based for. (User code will never be able to "observe" this flow condition, so I'm not sure why we need it.)

I still don't understand what the underlying purpose is that you need this flow condition for in the iterator check. Can you expand? (I want to make sure we're not dealing with an [XY problem](https://en.wikipedia.org/wiki/XY_problem).)

https://github.com/llvm/llvm-project/pull/80989